Design and synthesis of selective ligands for the FK506-binding protein 51

der Fakultät für Chemie und Pharmazie der Ludwig-Maximilians-Universität München

Design and Synthesis of Selective Ligands for the

FK506-binding Protein 51

Steffen Paul-Günter Gaali

aus

München, Deutschland 2012

Erklärung

Diese Dissertation wurde im Sinne von § 7 der Promotionsordnung vom 28. November 2011 von Herrn Prof. Christoph Turck betreut.

Eidesstattliche Versicherung

Diese Dissertation wurde eigenständig und ohne unerlaubte Hilfe erarbeitet.

München, 03.05.2012

... (Unterschrift des Autors / der Autorin)

Dissertation eingereicht am 03.05.2012

1. Gutachter: Prof. Christoph Turck

2. Gutachter: Prof. Roland Beckmann

Danksagung

An dieser Stelle möchte ich mich bei Prof. Dr. Dr. Florian Holsboer für die Möglichkeit bedanken, diese Arbeit am Max-Planck-Institut für Psychiatrie durchzuführen.

Besonders bedanken möchte ich mich bei Dr. Felix Hausch, der mir ermöglicht hat diese interessante und fordernde Doktorarbeit in seiner Forschungsgruppe durchzuführen. Seine hohe fachliche Kompetenz und die stete Diskussionsbereitschaft haben mir oftmals sehr weiter geholfen und ich habe viel für meine wissenschaftliche Zukunft gelernt.

Sehr herzlich möchte ich mich bei Prof. Turck und Prof. Beckmann bedanken für die Bereitschaft als Gutachter für meine Doktorarbeit zu fungieren. Des Weiteren bedanke ich mich bei Prof. Wanner, Prof. Heuschmann, Prof. Bracher und Prof. Hoffmann-Röder dafür als Prüfer zur Verfügung zu stehen.

Ich danke allen meinen Kollegen für das tolle Arbeitsklima und die Zeit die wir in und außerhalb des Labors miteinander verbracht haben. Besonderer Dank geht an die Chemie Crew von Labor 208 Christian Devigny und Yansong Wang für die tolle Zusammenarbeit und den Spaß im Labor, der das Arbeiten leicht gemacht hat und an meine persönliche Assay Abteilung Alexander Kirschner für das unermüdliche säen, ernten und mikroskopieren von verschiedensten Zellen, um meine Inhibitoren zu testen.

Ich bedanke mich bei Frau Dubler und Herr Dr. Stephenson von der NMR-Abteilung des Departments Chemie der LMU München.

Mein größter Dank gilt meiner Frau Maria sowie meinen Eltern, meinem Bruder und meinen Freunden ohne deren Rückhalt und Unterstützung in allen Lebenslagen ich nie so weit gekommen wäre.

Abstract

The large FK506-binding proteins FKBP51 and FKBP52 are Hsp90 associated cochaperones that modulate steroid hormone receptor signaling. It has been shown that FKBP51 is a negative regulator whereas FKBP52 is a positive regulator of the glucocorticoid receptor. A majority of patients suffering from depression show an altered response to glucocorticoids. Furthermore, polymorphisms in the FKBP51 encoding gene were associated with human stress response and several psychiatric disorders. Recently, FKBP51 knockout or knockdown was shown to have a protective effect on stress-coping behavior in animal models of anxiety and depression.

In a neuroblastoma cell line FKBP51 suppresses the elongation of neurites whereas FKBP52 enhanced it. All FKBP ligands reported so far, including rapamycin and FK506, show only negligible selectivity between FKBP51 and FKBP52, since the residues within the active site are completely conserved both on the sequence and the structural level. Due to the antagonistic effect of FKBP51 and FKBP52, the opposing activity of these proteins cannot be examined with the present FKBP inhibitors. Therefore, we envisioned a chemical genomics tool to address these selectivity problems. Using structure-based design and protein mutagenesis we engineered an enlarged cavity into the active sites of FKBP51 and FKBP52. In turn, we synthesized a series of complementary ligands with protruding side chains that were designed to fit into this new cavity and to prevent binding to the wild-type proteins. The best ligands of this series showed low nanomolar affinities while maintaining 500 to 1000-fold selectivity for mutated FKBP51/52 over wildtype proteins.

Using these artificially selective ligands in a cell model of neuronal differentiation (N2a cells), we showed that specific inhibition of overexpressed FKBP51 restores neurite outgrowth whereas specific inhibition of overexpressed FKBP52 has the opposite effect. This is the first proof of pharmacological activity of FKBP51 ligands in a relevant cellular model. Furthermore we unambiguously show that selectivity is crucial for the effect. This could at least in part explain the inconsistencies and conflicting results that have plagued the field of neuroimmunophilin FKBP ligands in the past.

During our synthesis campaign we made the discovery that certain ligands can induce a conformational change in the binding pocket of FKBP51 and that these substances consistently show substantial selectivity versus FKBP52. Based on several co-crystal structures we rationally designed a series of these induced fit ligands which finally led to inhibitors (iFit-1, IFit-2) with low nanomolar affinities (4-6 nM) for wildtype FKBP51 and up to 10000 fold selectivity versus FKBP52. These ligands are the most potent and selective ligands reported for FKBP51 so far. In a neurite outgrowth assay they enhanced neurite outgrowth whereas FK506 was less active. These ligands provide the basis for the development of drug-like FKBP51 inhibitors to pharmacologically probe the role of FKBP51 in a whole animal context.

A. Introduction ... 1

1. The biology of Immunophilins ... 1

1.1 FK506-binding proteins and Cyclophilins ... 1

1.2 The biology of FKBP51 and FKBP52 ... 4

1.2.1 Structure and function of FKBP51 and FKBP52 ... 4

1.2.2 The role of FKBP51 and FKBP52 in steroid receptor signaling ... 5

1.2.3 Effects of FKBP51 and FKBP52 on the endocrine system ... 7

1.2.3.1 FKBP52 knockout mouse ... 7

1.2.3.2 FKBP51 knockout mouse ... 7

1.2.4 FKBP51 in stress related diseases ... 8

1.2.5 Cancer and cell proliferation ... 9

1.2.6 Immune function ... 10

1.2.7 Effect on neurodegenerative diseases ... 10

1.3 Chemical biology of FKBP ligands ... 11

1.3.1 Immunosuppressive FKBP ligands ... 11

1.3.2 Non-immunosuppressive FKBP ligands ... 12

1.4 Targeting the PPIase binding pocket of FKBP51 and FKBP52 ... 15

1.5 Artificially selective ligands ... 19

2. Aim of the study ... 22

B. Results and Discussion ... 23

1. Chemical genomics to selectively address FKBP-sub-types ... 23

1.1 Design of FKBP mutant specific engineered (FMSE) ligands ... 23

1.2 Synthesis of the FMSE ligands ... 24

1.3 Biochemical characterization of the FMSE-ligand-mutant pairs ... 27

1.4 Effect of selective inhibition of mutated FKBP51 and 52 on neurite outgrowth in N2a

neuroblastoma cells ... 31

2. Solving the selectivity issue by an induced fit mechanism ... 35

2.1 Induced fit as a basis for selectivity ... 35

2.2 Synthesis of iFit FKBP51 ligands ... 38

2.2.1 Cyclopropylmethyl and benzyl series ... 39

2.2.1.1 Design and synthesis of Cα cyclopropylmethyl and benzyl ligands ... 39

2.2.1.2 Biochemical activity of cyclopropylmethyl and benzyl ligands ... 41

2.2.2.1 Design and synthesis of cyclohexenyl/cyclohexyl ligands ... 45

2.2.2.2 Biochemical activity of cyclohexenyl/cyclohexyl ligands ... 47

2.2.2.3 Effect of iFit ligands on neurite outgrowth ... 52

2.2.3 Cα-hydroxy series ... 53

2.2.4 Cα symmetric ligand series ... 55

2.2.5 Synthesis of Cα substituted rapamycin ... 56

2.3 Structural basis for the selectivity ... 58

2.3.1 Structure-activity relationship of the iFit ligands ... 58

2.3.2 Quantification of the induced conformational change ... 60

2.3.3 Evaluation of the co-crystal structures ... 61

2.3.3.1 The Cα-Substituent... 61

2.3.3.2 The Trimethoxyphenyl Moiety ... 62

2.3.3.3 The Top Group ... 63

3. Summary and Outlook on Selective FKBP Ligands ... 65

4. Fluorescent Immunophilin Tracers ... 67

4.1 Synthesis of fluorescent rapamycin derivatives ... 67

4.2 Synthesis of a fluorescent iFit ligand ... 69

4.3 Facile synthesis of a fluorescent CsA analogue to study Cyclophilin 40 and Cyclophilin

18 ligands ... 70

4.3.1 Synthesis of the tracer ... 71

4.3.2 Development of a fluorescence polarization assay for cyclophilin 40 and Cyp18 ... 72

C. Experimental Section ... 74

1. Analytical Methods ... 74

1.1 Nuclear magnetic resonance ... 74

1.2 Mass spectroscopy ... 74

1.3 HPLC ... 74

1.4 Silica chromatography ... 76

1.5 Data analysis of neurite outgrowth ... 77

2. Reagents and solvents ... 77

2.1 Reagents ... 77

2.2 Non-commercial reagents... 80

2.3 Solvents ... 80

4. Synthesis of used compounds ... 81

4.1 General synthesis procedure A for the coupling of morpholine containing top-groups

... 81

4.2 General synthesis procedure B for the coupling of free acid top-groups ... 82

4.3 Synthetic procedures ... 82

4.4 Biochemical Methods ... 147

D. Abbrevations ... 149

E. Annex ... 151

F. References ... 155

G. Curriculum Vitae ... 164

1

A. Introduction

1. The biology of Immunophilins

1.1 FK506-binding proteins and Cyclophilins

Fig. 1: The prototypic FKBP ligands FK506 and rapamycin and the cyclophilin ligand CsA. Blue (FK506, rapamycin): FKBP binding domain, blue (CsA) cyclophilin binding domain. Green (CsA, FK506) calcineurin binding domain, green (rapamycin) mTOR binding domain.

FK506-binding proteins (FKBP) and cyclophilins (Cyp) belong to the class of immunophilins which are defined by their ability to bind immunosuppressive ligands like FK506, rapamycin (Rap), and Cyclosporin A (CsA, Fig. 1). FKBPs and Cyps minimally contain a peptidyl-prolyl-isomerase (PPIase) domain that catalyzes the interconversion of cis-trans isomers of X-Pro peptides and that binds to immunosuppressive drugs. For all aminoacids except proline the equilibrium of cis/trans isomerization lies on the trans side. In proteins these 19 aminoacids adopt almost exclusively the trans configuration. X-Proline dipeptides can occur either in the cis or the trans configuration in folded proteins, whereas in the unfolded state there is an ratio of cis/trans 2:8. Therefore in many protein folding processes the cis/trans isomerization of X-Pro displays the rate determining step. PPIases help proteins to fold in a correct way by catalyzing the isomerization of proline residues.1 The immunosuppression is not mediated by inhibition of the PPIase activity but by enabling FKBPs to form a ternary complex with calcineurin (for FK506) or mTOR (for Rap). In complex with FKBP-FK506 calcineurin is not able to dephosphorylate nuclear factor of activated T-cells (NF-AT) which is needed for IL-2 expression and T-cell activation. The main mediators are FKBP12, and partially FKBP12.6 and

2

FKBP51.2 mTOR assembles two complexes mTORC1 and mTORC2. The kinase activity of mTORC1 can be specifically inhibited by FKBP-Rap complexes which in turn leads to less phosphorylation of p70S6Kinase and 4E-BP1, both key regulators in protein translation and thereby causing the immunosuppressive effect.2, 3,4

Fig 2: Proline cis/trans isomerization

The human genome encodes 17 different FKBPs, which are named according to their size in kilodaltons (e.g., FKBP12, FKBP38, FKBP51 and FKBP52). Table 1 shows the different known FKBPs and their biochemical roles in mammalian cells known so far5.

Name

Partners Functions Cellular Compartment hFKBP12a6, 7 hFKBP12.68, 9 hFKBP12c5 FK506/Calcineurin Rap/mTOR Type I TGFβ receptor Muscle ryanodine receptor Inositol receptor

cardiac ryanodine receptor

regulator of cell cycle Cytosol

hFKBP15p5 Protein coding cofactor ER

hFKBP22p5 hFKBP25p5 hFKBP24p5 hFKBP63p5

hFKBP65p10 elastin chaperone

hFKBP3611, 12 clathrin and Hsp72 glyceraldehyde-3-phosphate dehydrogenase inhibitor

Nuclear

hFKBP3713 Aryl receptor Transcription of genes Cytosol

hFKBP37i14 Amarosis syndrome Cytosol

3

Hedgehog signalling

hFKBP5116 SHR, HSP90, Akt Negative Modulator of SHR Cytosol

hFKBP5217 SHR, HSP90 Positive Modulator of SHR Cytosol

hFKBP25n18 YY1, HMG-II Transcription of genes Nuclear/Cytosolic

hFKBP1355 F-Actin Colocalized with F-Actin in

growth cones of dorsal root ganglion neurons

Cytosol

Table 1: Biochemical roles and distribution of human FKBPs

The second class of immunophilins are the cyclophilins. The human genome encodes at least 16 unique cyclophilins, all containing a highly conserved Cyp18-homology domain, which shows PPIase activity. Many of them bind tightly to the unselective cyclophilin ligands CsA and sanglifehrin A. 19, 20 The Cyp-CsA complex forms a ternary complex with the phosphatase calcineurin (CN) similar to FKBP12-FK506. In this heterocomplex calcineurin is also unable to dephosphorylate its substrate NF-AT which is required for T-cell activation which again leads to the immunosuppressive effect3.

4

1.2 The biology of FKBP51 and FKBP52

1.2.1 Structure and function of FKBP51 and FKBP52

Fig 3: Crystal structures of FKBP51 and FKBP52

The large FKBP homologs FKBP51 and FKBP52 have first been identified in complex with steroid hormone receptors (SHR).21, 22 The binding to SHRs is mediated by the heatshock protein 90 (Hsp90), where they act as co-chaperones.23 Since then these proteins received great attention because of their steroid hormone signaling-regulating roles. Many endocrine related diseases are known for which FKBP51 and FKBP52 are potential therapeutic targets, such as for example stress related diseases, prostate cancer, breast cancer, male and female contraception and metabolic diseases. To better understand the role of these FKBPs in these diseases new non-immunosuppressive ligands are needed.

FKBP51 and FKBP52 are close homologs and share 70% sequence similarity.23 They possess a similar domain architecture (Fig. 3), consisting of the FKBP12 like N-terminal PPIase domain (FK1), followed by another FKBP12 like domain (FK2) which although structurally similar to FK1 possesses no PPIase activity. At the C-terminus a tetratricopeptide (TPR) domain facilitates the binding to the EEVD motif at the C-terminus of Hsp90.24, 25 The overall architecture of the domains of FKBP51 and FKBP52 is very similar. The orientation of FK1 and FK2 differ only slightly but the TPR domain orientation of FKBP52 is tilted compared to that of FKBP51. It has to be considered that FKBP52 was not crystalized

TPR domain

FK2

FK1

FK2

FK1

FKBP51

FKBP52

5

as a whole, but in two parts due to its instability in solution. The crystal structure in Fig. 3 shows the reconstruction based on the two parts. The loop linking the FK2 and the TPR domain could be flexible and the orientations of the TPR domains could be due to different crystal packing.

The FK1 and FK2 domain are connected by a linker of seven to nine amino acids. FKBP52 has a casein kinase 2 phosphorylation (CK2) sequence (TEEED) in this linker which is not present in FKBP51 (the correspondent sequence is FED). It is thought that the phosphorylation of FKBP52 by CK2 at T143 decreases binding to Hsp90 and thereby abrogates the activating effect. This effect could be due to a reorientation of the FK1 domain upon phosphorylation.26

1.2.2 The role of FKBP51 and FKBP52 in steroid receptor signaling

FKBP51 and FKBP52 are regulators of steroid hormone receptor (SHR) binding activity. In most reports FKBP51 acts as a negative modulator on SHRs27_{, whereas FKBP52 is a positive regulator of} androgen receptor (AR)28, glucocorticoid receptor (GR)29 and progesterone receptor (PR)30. Fig. 4 shows a model of the maturation and regulation of SHRs. Either FKBP51 or FKBP52 enters the mature Hsp90-dimer-SHR complex, which is stabilized by p23. The FKBP binds to the C-terminus of Hsp90 via the TPR domain. The present model proposes the FK1 domain and especially the proline rich loop of FKBP51 and FKBP52 interacts directly with the ligand binding domain of the SHR. If FKBP51 is present the binding affinity for the respective hormone decreases, whereas if FKBP52 is in the complex the binding affinity is increased.31

6

Fig. 4: Model of FKBP51 and 52 on steroid hormone maturation and ligand binding.

The PPIase enzymatic activity is not required for the modulation of the SHRs, but the FK1 domain and especially the proline rich loop which sits on top of the binding pocket, is crucial.32 Differences in this loop seem to be the cause for the different functions of the FKBPs, shown by the mutations, A116V and L119P in FKBP51 that switched the activity to full FKBP52-like characteristics towards AR activation.32

FKBP51 and FKBP52 also play a role in steroid hormone receptor localization. In the ligand free state the SHRs primarily stay in the cytoplasm, whereas ligand bound SHRs are mainly nuclear or translocate to the nucleus.33, 34 It has been suggested that the accumulation of ligand bound SHR in the nucleus is enhanced by active retrograde transport driven by the dynein-dynactin complex which co-immunoprecipitate with the Hsp90-FKBP52 and with the GR and MR.35, 36

7

1.2.3 Effects of FKBP51 and FKBP52 on the endocrine system

1.2.3.1 FKBP52 knockout mouse

FKBP52 knockout mice (52KO) are viable but females are completely infertile. Male 52KO mice display a phenotype consistent with androgen insensitivity syndrome, hypospadias, penis length and weight of the penis was reduced, smaller seminal vesicles, smaller prostate glands, slightly lower sperm motility, collectively showing that mainly secondary sex organs are affected whereas primary sex organs like testes seemed to be unaffected.28, 37 Female 52KO mice show no big change in phenotype but are sterile. This is due to progesterone insensitivity causing failures in decidualization and embryonic implantation.38 Thus, FKBP52 is crucial for correct development of reproductive organs in male and female mice which is mainly caused by AR and PR insensitivity.

1.2.3.2 FKBP51 knockout mouse

Under basal conditions FKBP51 knockout mice (51KO) show no robust phenotype. 51KO male and female mice are fertile and males show normal reproductive organs. Thus AR signaling is unaffected and also no changes in GR activity could be observed.37 A possible explanation for the unanticipated absence of an effect on GR is the nature of the cortisol secretion under stress, and indeed recently Touma and coworkers could show, that 51KO in mice leads to a more active coping behavior after exposure to different types of stress. Additionally the hypothalamus-pituitary-adrenal (HPA) axis response on stress was altered. 51KO mice showed a stronger suppression of corticosterone secretion after treatment with a low dose of dexamethasone.39 These findings were supported by the results of Hartmann et al. who showed in a chronic model of social defeat stress, that 51KO mice responded less to a novel acute stimulus and showed an enhanced recovery, as well as more active stress-coping behavior.40 Additionally O’Leary and coworkers demonstrated that FKBP51 deficiency in aged mice led to more active stress-coping in the forced swim test and the tail suspension test. Both are well established paradigms to assess antidepressive effects.41 All these findings strongly support the hypothesis that FKBP51 plays an important role in endocrine regulation of the HPA axis by reducing GR responsiveness. This makes FKBP51 a promising target in stress related diseases.

8

1.2.4 FKBP51 in stress related diseases

In stress related diseases such as major depression, bipolar disorder, post-traumatic stress disorder (PTSD) and anxiety disorder patients often display an imbalance in the stress hormone system called the hypothalamus-pituitary-adrenal (HPA) axis (Fig. 5). In healthy individuals this hormone system triggers the physiological and behavioral response to stressors. This can be measured by an increase in blood cortisol levels that peaks after 15-30 min and then slowly declines after termination of the stressor. Cortisol is a catabolic steroid hormone that activates energy metabolism in various tissues and acts as a negative regulator on the HPA axis.42

Fig. 5: Hypothalamus-pituitary-adrenal axis with hormone regulation cascades

Upon stress the hypothalamus secretes corticotropin releasing hormone (CRH) which induces the production of adrenocorticotropic hormone (ACTH) in the pituitary gland. ACTH in turn increases the release of cortisol in the adrenal gland into the blood. Cortisol is binding to the GR and MR which in turn inhibit the further release of CRH and ACTH thereby maintaining homeostasis of the HPA axis. Additionally, an ultrashort feedback loop is thought to be present at the cellular level. Activated GR

9

increases the FKBP51 expression which in turn decreases the affinity of the GR for cortisol. Malfunctions in these negative feedback loops are thought to be a cause for an inappropriate reaction of the HPA axis to stress which is often observed in depressive patients.

FKBP51 is a known negative modulator of GR activity. Its physiological relevance was supported by findings in squirrel monkeys which show an increased blood cortisol level associated to decreased GR activity and an overexpression of the more potent squirrel monkey FKBP51.43, 44 _{These findings} initiated human genetic studies on FKBP51 in major depression. In these studies identified single nucleotide polymorphisms in the FKBP51 encoding gene were associated with the response to antidepressants and with more lifetime depressive episodes.45 Similar studies followed that confirmed these findings46, 47 and found gender-specific effects.47, 48 FKBP51 genetic variants could also be linked to bipolar disorders48 and significant associations where also found to suicidal events.49-52 Polymorphisms in the FKBP51 encoding gene also influence the recovery from psychosocial stress in healthy individuals53. Another link could be observed from FKBP51 gene variants to peritraumatic dissociation54 which is an important risk factor for development of a PTSD.55 The connection to PTSD was also found in other studies.56 All these findings clearly show that FKBP51 contributes to the etiology of stress-related psychiatric disorders.

1.2.5 Cancer and cell proliferation

FKBP51 is up-regulated by androgens (natural: dehydrotestosterone, synthetic: R1881) which made it an interesting target for androgen dependent cancer types. Indeed, FKBP51 has consistently been reported to be up-regulated in human prostate cancer cells.57 FKBP51 was also found to be up-regulated in prostatic hyperblasia.58 Further FKBP51 was shown to promote the assembly of the Hsp90 chaperone complex and thereby regulates androgen receptor signaling in prostate cancer cells.59, 60 However, the unanticipated potentiation of AR by FKBP51 is a very special case because in all other reported studies FKBP51 is a negative regulator of SHR. Although, this effect was not seen in all reported studies and seems to be cell-type dependent.61 The FKBP unselective ligand FK506 was shown to inhibit cell growth after androgen stimulation in a new prostate cancer type where FKBP51 and FKBP52 are overexpressed.

It was demonstrated that FKBP51 suppresses the proliferation of colorectal adenocarcinoma, possibly due to its deactivating effect on glucocorticoid receptors62. Following dexamethasone treatment myeloma cells show prior cell death up-regulation of FKBP51. This could be exploited to enhance the myeloma killing effect of dexamethasone in future63. By an siRNA approach a link of FKBP51 to drug-induced NF-κB activation in human acute lymphoblastic leukemia could be shown.

10

This was supported by inhibition using rapamycin64. In a recent study the role of FKBP51 in melanocyte malignancy was outlined65. In a different cell line Pei and coworkers were able to

identify FKBP51 as a negative regulator of the Akt pathway by serving as a scaffolding protein for PHLPP. A reduced expression of FKBP51 in certain cancer types could be correlated with increased AKT phosphorylation which resulted in a reduced cell sensitivity to chemotherapeutics66.

Much less is known about the implications of FKBP52 in cancer but recently it could be shown that prevention of hormone-induced dissociation of the Hsp90-FKBP52-AR complex results in inhibition of androgen-stimulated prostate cancer cell proliferation67.

Due to their modulating actions on steroid hormone receptors and their implications in the according diseases, FKBP51 and FKBP52 represent promising drug targets for anti-cancer therapy.

1.2.6 Immune function

FKBPs also play a role in immune function and inflammation. The best known effect of FKBPs is their ability to form ternary complexes with the immunosuppressive drugs FK506 and rapamycin (see

Chapter 1.1). Besides these prominent immunosuppressive effects various other implications were

published during the last years. Recent studies showed FKBP51 to be up-regulated in CD34+ bone marrow cells in patients with rheumatoid arthritis.68, 69 Park and co-workers demonstrated that FKBP51 can modulate NF-κB-dependent gene expression in Newcastle disease virus-infected chickens.70 _{Further it was shown that FKBP51 modulates the stability of IκB and the phosphorylation} of NF-κB and enhances its DNA binding.71 A very recent study connects FKBP51 expression with asthma after administration on inhaled corticosteroids.72 Additionally, in patients suffering from chronic obstructive pulmonary disease an up-regulation of FKBP51 could be observed.73 FKBP51 plays also a role in endogenous MHC class II-restricted antigen presentation. FK506 was able to inhibit the presentation of endogenous MHC class II-restricted minor histocompatibility antigens in primary dendritic cells (DC) in vitro. This effect could be rescued by RNAi mediated reduction of FKBP51.74

1.2.7 Effect on neurodegenerative diseases

Besides their role in immunosuppression, FKBPs have repeatedly been linked to neurodegeneration in several animal models like transient focal cerebral ischemia in rats or in MPTP mouse models of Parkinson's disease.75, 76 Gold et al. showed in a rat sciatic nerve crush model that the immunosuppressive drug FK506 accelerated nerve (re-)generation.77 This effect could also be

11

transferred to human patients receiving hand transplants.78 FK506 binds unselectively to FKBP proteins and the mediator of the the neuroprotective effect could not be elucidated so far. Early research was focused on the most abundant FKBP12 as the main mediator of the neuroprotective effect of FK506. It could be shown that adding FK506 to primary neurons during or after glucose deprivation limited the induced damage. This effect could be reverted by a FKBP12 antibody or the competitive inhibitor rapamycin.79 _{Importantly, the use of non-immunosuppressive derivatives of} FK506 which are not able to form a ternary complex ruled out a calcineurin-dependent mechanism for the observed neuroprotective activity.80, 81 In primary hippocampal cell cultures from FKBP12 knockout mice FK506 retained neurotrophic activity, thus devalidating the prototypic FKBP12 as the relevant target in this model. The mitochondrial FKBP38 and the Hsp90 co-chaperone FKBP52 have been suggested as alternative targets82, 83 (for detailed reviews see84,85). Present FKBP ligands that show neuroprotective effects bind unselectively the whole group of FKBPs, with most research focused on FKBP12, FKBP38 and FKBP52. Quinta and co-workers could show that neurite outgrowth in mouse N2a cells is favored by FKBP52 over-expression or FKBP51 knock-down, and is impaired by FKBP52 knock-down or FKBP51 over-expression, nicely showing the antagonistic activities of FKBP51 and 52 on neuronal differentiation.86 FKBP51 and FKBP52 were also found to play a role in tau turnover which is a key phenomenon in Alzheimer’s disease.87, 88

All these findings suggest that immunophilins and especially the larger homologs FKBP51 and FKBP52 are important for neuronal processes involved in neurprotection, neuroregeneration and neuronal differentiation. Additionally, because of the antagonistic effects of FKBP51 and FKBP52 in different systems it is particularly important to develop FKBP subtype selective Ligands to dissect the opposing roles of these proteins.

1.3 Chemical biology of FKBP ligands

1.3.1 Immunosuppressive FKBP ligands

Since the discovery of FK506 and rapamycin (Fig. 1) in the 1990s and the characterization of their immunosuppressive effect a lot of research has been devoted to the improvement of these ligands in terms of side effects, solubility and efficacy. Fig. 6 shows FK506 and rapamycin analogs that are used in the clinic. The efforts in this field increased even more after the discoveries that rapamycin has beneficial effect on longevity in mice89, improves behavioral and cognitive deficits in models of

12

neurodegeneration like Alzheimer’s90, Parkinson’s91 and Huntington’s disease92 and in cancers due to misregulated mTOR pathway.93

Fig. 6: Clinically used FK506 and rapamycin analogs84

1.3.2 Non-immunosuppressive FKBP ligands

Up to now almost no drugs are available to treat chronic neurodegenerative diseases. Gold and co-workers found in the early 1990s that besides its immunosuppressive effects FK506 has also neurotrphic activity. These effects were shown in a rat sciatic nerve model.77, 94 These findings stimulated a whole field to search for non-immunosuppressive immunophilin ligands that still display this neuroprotective effect. These ligands were termed neuroimmunophilin ligands.

Almost 20 years of medicinal chemistry and biochemistry efforts produced a variety of non-immunosuppressive ligands based on the known natural products (Fig. 7) where the effector domain is changed. This abolished the binding to calcineurin/ mTOR (e.g., FK1706, meridamycin, normeridamycin, ILS920, Way-124466, Wye-592, L685-818). These ligands demonstrated their effect in animal models of cerebral ischemia83, 95, traumatic brain injury96, diabetic neuropathy97, Parkinson’s disease75, 98, 99, and various types of physical neuronal injury.81, 100-102

13

Fig. 7: Neuroimmunophilin ligands based on biosynthetic or semi-synthetic analogs of

FK506 or rapamycin84

In addition a vast number of synthetic FKBP ligands have been reported which are based on the pipecolyl/prolyl diketoamide core derived from FK506 and rapamycin. All these compounds lack the effector domain of FK506 and are thus not immunosuppressive. Fig. 8 shows a collection of reported synthetic neuroimmunophilin ligands that showed neurotrophic activity. VX-10,367 and VX-7109 where patented for stimulating neurite growth in nerve cells and are the most potent FKBP12 ligands known to date.103 GPI1046 received a lot of attention due to its effect on neurite outgrowth from sensory neuronal cultures with reports of picomolar potency In vivo. Additionally GPI-1046 stimulated the regeneration of lesioned sciatic nerve axons.80, 104 Analogs of GPI1046 were also published to be neurotrophic (GPI1485, JNJ460).105 However these results were challenged by other

14

groups.104, 106 Moreover, GPI1046 was also reported to be inactive in PPIase assay by us and others.85 Our own unpublished data shows that GP1046 is also inactive for the higher homologs FKBP51 and FKBP52.

Most of the studies in the literature are focused on FKBP12 but Gold and coworkers showed that also other proteins can mediate the effect.82 This was corroborated by using various FK506 analogs that are claimed not to bind FKBP12 (VX-853, V-13,661 and V-13,670, Fig. 8).75, 102 _{Furthermore the} selective FKBP38 inhibitor DM-CHX was shown to be active in an animal model of focal cerebral ischemia.83

All these results show that FKBP ligands can have neuroprotective or neurotrophic activities and may be potentially useful in certain neurodegenerative diseases or after neuronal loss. Although a lot of inconsistencies still exist possibly by differences in cellular and animal models or ligands used. Also the relevant targets are still controversially discussed.85

Fig.8: Synthetic neuroimmunophilin ligands. The core of FK506 or rapamycin or equivalent groups

15

1.4 Targeting the PPIase binding pocket of FKBP51 and FKBP52

Fig. 9: (A) Natural product FKBP ligand FK506, FKBP binding domain (blue), effector domain (green). (B) Synthetic FKBP ligand SLF FKBP binding domain (blue). (C) Co-crystal structure of FK506 and FKBP51. (D) Co-crystal structure of SLFand FKBP51.

As outlined in the previous chapter most of the synthetic immunophilin ligands were designed to bind FKBP12. These ligands were derived from the diketoamide pipecolinic core of FK506 and rapamycin lacking the effector domain and are exemplified by SLF107 (Fig. 9).The pyranose ring was exchanged by a tert-pentyl group which proved to be a good isoster. SLF showed binding affinity for FKBP12 in the range of FK506 (low nanomolar) but for the larger homologs FKBP51 and FKBP52 it was substantially less affine (low micromolar).108 Therefore, the Hausch group solved the co-crystal structure of SLF and FKBP51109 and compared it to the co-crystal structure of FK506 and FKBP51110 as a starting point for rational ligand design. The amino acids of the binding pocket are almost superimposeable and SLF also shows the important hydrogen bonds from I87 to the pipecolate carbonyl group and from Y113 to the amide carbonyl group. W90 forms the bottom of the binding pocket and the pipecolyl ring sits on top of the indole ring. SLF was the only ligand in the literature

W90

I87

Y113

W90

I87

Y113

(A)

(B)

(C)

(D)

16

that was described for FKBP51108 at the start of this thesis while for FKBP52 three ligands were known.95, 105

Gopalakrishnan and co-workers performed the first structure activity relationship (SAR) analysis to determine the contributions of individual substructures of SLF (Fig. 10).109 The affinities were measured using the fluorescence polarization assay described by Kozany et al.108 SLF bound to FKBP51 with ~8µM and to FKBP52 with ~10µM.

Fig. 10: Synthetic analogs of SLF for SAR analysis

Replacement of the pipecolyl core by proline or 4,5-dehydropipecolinic acid resulted in a 4-6 fold reduction in potency. The change to thiomorpholine-3-carboxylic acid abolished binding to FKBP51 and FKBP52.

Furthermore they employed different top-groups. The smaller groups (Fig. 10, A-D) showed no binding to FKBP51 and FKBP52. To eliminate the free charge at the free acid moiety of F and G they changed it to morpholine H which increased the binding affinity by 2-4 fold compared to SLF and a slight preference for FKBP52 could be observed. This trend could also be seen in a sulfonamide series of compounds also published by the Hausch group111. They replaced the ester at C1 G by an amide which abolished binding to FKBP51 and FKBP52. Finally, they replaced the oxyacetyl group of the SLF top-group by an amine I which resulted in the best binding compound 1 (Fig. 11) of this series. It showed binding to FKBP51 of ~4 µM and to FKBP52 of ~1µM.

17

Fig. 11: Exchange of tert-pentyl by 3,4,5-trimethoxyphenyl

They continued by replacing the tert-pentyl group with 3,4,5-trimethoxyphenyl (Fig. 11) which resulted in a 2 fold decrease in binding for the larger FKBPs. Additionally two clinically used non-immunosuppressive FK506 analogs were tested on their binding to FKBP51 and FKBP52 (Fig. 8). Biricodar showed affinity in the range of SLF ~8 µM whereas Timcodar showed no affinity for any FKBP tested.

Fig. 12: Overlay of the important amino acids of the binding pocket of FKBP12, FKBP51 and FKBP52. The not conserved

amino acids are marked in red.

FKBP12 Y26 F36 D37 R42 N43 F46 E54 V55 I56 W59 Y82 H87 P88 I91 F99 FKBP51 Y57 F67 D68 R73 N74 F77 Q85 V86 I87 W90 Y113 S118 L119 I122 F130 FKBP52 Y57 F67 D68 R73 K74 F77 E85 V86 I87 W90 Y113 S118 P119 I122 F130 W90

40s loop

proline rich

loop

18

The binding pocket of all FKBPs is highly conserved .The most prominent changes are found in the 70s loop (amino acids 71-76 of FKBP51/52) and the proline rich loop (amino acids 118-122 of FKBP51/52) (Fig. 12). The most important change in the amino acid sequence between FKBP51 and FKBP52 can be found in position 119 in the proline rich loop. A L119P mutation in FKBP51, which introduces the proline found in FKBP52, conferred significant potentiation activity towards steroid hormone receptors, whereas the converse P119L mutation in FKBP52 decreased potentiation.32 Thus, they planned to target the proline rich loop with with a new series of compounds comprising a substituted cyclohexyl ring (Fig. 13) instead of the tert-pentyl. This would be more close to the pyranose ring of FK506.

Fig. 13: Cyclohexyl substituted ligand series targeting the proline rich loop

The SAR of these compounds and the crystal structures that they published showed that FKBP51 and FKBP52 are tolerant to different stereochemistries at the cyclohexyl substituent. The best binding compounds of this series C1 and C2 (Fig. 14) show binding affinities of 1 µM to 4 µM.

19

For FKBP12 it was shown that the diketo amide moiety can be bioisosterically exchanged to sulfonamide112, 113. To determine the binding of sulfonamide ligands to FKBP51 and FKBP52 Gopalakrishnan et al. developed a solid phase strategy for the synthesis of a focused sulfonamide library111.

Out of 36 compounds with aromatic sulfonamides they identified 5 hits for the binding to FKBP51 and FKBP52 which displayed a slight preference for FKBP52. The hits showed moderate binding affinities with ~10 µM.

For the best hits the morpholine top-group (Fig. 15) was employed and increased the binding affinity to nanomolar levels for S1 and to low micromolar for S2. Therefore ligand S1 displays the best known ligand for the large FKBPs to date. With this strategy they could show that a bioisosteric replacement of the diketo amide to sulfonamide with conservation of the hydrogen bonds leads to potent FKBP51 and FKBP52 inhibitors.

All of the described ligands in this chapter unfortunately show no selectivity between FKBP51 and FKBP52 and at least 10 to 100-fold higher affinity for FKBP12.

Fig. 15: Top-group substitutions of the best hits

1.5 Artificially selective ligands

20

In the early 1990s Spencer et al.114 developed chemical inducers of dimerization (CID) to control intracellular signal pathways that are normally controlled by protein-protein interactions. Therefore, they conjugated two FK506 molecules together via a linker and termed it FK1012. This was improved by the development of the bivalent synthetic analog AP1510115 (Fig. 16). These CIDs can bind and dimerize proteins of interest which are fused to a FKBP12 “tag” and thereby specifically activate signaling. The major disadvantage is the high affinity to endogenous FKBPs which are highly expressed. This leads to unwanted heterodimers that interfere with the signaling pathways which are to be observed.

Fig. 17: Priciple of artificial selective ligands. Bulky

modification at the ligand abolishes binding to the wildtype but allows binding to the mutant

To address this selectivity problem the group of Holt116 used a chemical genomics approach to redesign the FKBP12-ligand interface by engineering a new pocket into the active site. At the same time they synthesized ligands that exploit the newly formed cavity in the binding pocket (Fig. 17 and

Fig. 18). These ligands showed a substantial decrease in binding affinity to the wildtype but high

21

Fig. 18: (A) Model of ligand 10 in the crystal structure of FKBP12WT. Steric clash of the Cα substituent with Phe36 (red). (B) Co-crystal structure of 10 with the mutated FKBP12F36V. Cα substituent fits into the new formed cavity.

To achieve that goal they investigated the co-crystal structure of FK506 and FKBP12 and concluded that an exchange of the Cα carbonyl to larger substituents would sterically clash with either Tyr26 or Phe36 and should abolish or decrease binding. In turn a compensating mutation at one of these amino acids would restore binding (Fig. 17 and Fig. 18). The best ligand (1, Fig. 20) showed an affinity of 1.8 nM to the F36V mutant of FKBP12 and to the wildtype FKBP12 of 2930 nM. This ligand can differentiate 1,000-fold between WT and was further dimerized via a short linker to form AP20187 (Fig. 19) which was then used to activate Fas signaling in a mouse model of conditional cell ablation116.

Fig. 19 Mutant selective FKBP12 CID AP20187.

CIDs have been used in a broad range of applications for dimerization of for example membrane receptors: Erythropitin receptor117, PDGF-ß-R / Insulin receptor118, epithelial growth factor receptors / hepatocyte growth factor / thrombopoietin receptor119 or for the induced activation of apoptosis by dimerization of the FAS receptor or the dimerization of caspases115, 120-122.

Banaszynski et al. from the Wandless group further expanded that field by designing a method to reversibly regulate protein stability in living cells using a synthetic analog termed Shield-1123

Phe36

Val36

22

Another very prominent example of the bump and hole strategy is the selective inhibition of mutant kinases shown by Bishop from the lab of Kevan Shokat.124 With this technique they designed a selective cdc28 inhibitor and showed for the first time the effect of the inhibition of this specific kinase. This technique was also applied to other kinases.125

2. Aim of the study

Fig. 20: Starting point of the synthetic campaign

FKBP51 is a known negative modulator of glucocorticoid receptor (GR) activity whereas its closest homolog FKBP52 activates the GR.

The natural product FK506, rapamycin and all other reported FKBP ligands show if at all a selectivity for FKBP12 because of their high structural and sequence similarity84. Due to the opposing effects of FKBP51 and FKBP52 these ligands are not suited to study the role of these proteins in GR signalling. The goal of the study was to design and synthesize ligands that solve that selectivity issue of the known FKBP ligands.

We envisioned a chemical genomics tool to artificially design selective ligand mutant pairs for FKBP51 and FKBP52. Therefore structure-based design should be used to synthesize ligands with protruding Cα substituents and site directed mutagenesis to introduce an enlarged cavity into the active sites of FKBP51 and FKBP52 to compensate for the Cα substituent. Starting point for the synthesis was the FKBP12F36V mutant selective ligand 1116 (Fig. 20). This chemical genomics tool was intended to be applied to more complex in vitro GR binding assays and cellular assays like GR reportergene assays to probe the pharmacological tractability of FKBP51 and its potential as a druggable target.

23

B. Results and Discussion

1. Chemical genomics to selectively address FKBP-sub-types

1.1 Design of FKBP mutant specific engineered (FMSE) ligands

Fig. 21: FKBP mutant-specific engineered ligand 1

All FKBP ligands reported so far, including rapamycin and FK506, show only negligible selectivity for individual FKBP homologs since the residues within the active site are completely conserved both on the sequence and the structural level except for FKBP38.110 FKBP51 and 52 have been shown to have opposing effects on steroid hormone receptors as well as on neurite outgrowth. FKBP51 in most cases reduces receptor sensitivity, whereas FKBP52 is a positive regulator of SHRs.31 Likewise in a neuroblastoma cell line FKBP51 suppressed the elongation of neurites whereas FKBP52 enhanced it.86 Due to these antagonistic effects, these proteins cannot be examined with present FKBP inhibitors. Chemical genomics provides for this case the perfect tool to artificially overcome this selectivity issue by engineered mutant-ligand pairs. A large hydrophobic amino acid in the active site is mutated to a smaller amino acid, which generates a new hole in the binding pocket. In turn, a complementary ligand is engineered with a protruding sidechain that fits into this new cavity. This sidechain performs two tasks; first it should increase the affinity to the mutated protein and second decrease the affinity to the wildtype. In previous work in the Hausch lab phenylalanine 67 was mutated to valine to open a new cavity in the binding pocket. Complementarily, compound 1

24

(Fig. 21) with a Cα ethyl in (S)-configuration was synthesized, which is a tight and selective binder of FKBP12 carrying a homologous F36V mutation.116 Compound 1 showed moderate binding affinity to FKBP51F67V (51mut) of IC50~2µM and to FKBP52F67V (52mut) of IC50~22µM. The affinities to the wildtype proteins were >100 µM. While this compound showed good selectivity vs the wildtype proteins of greater than 50 fold, the affinity had to be improved substantially to allow for cellular experiments. We thus used 1 as a starting point to synthesize a series of analogs to optimize the interaction with the mutated FKBP51/52 binding site (Fig. 22).

Fig. 22: General structure of FKBP mutant specific engineered

ligands

1.2 Synthesis of the FMSE ligands

The top group of the ligands 5a was synthesized by an improved procedure based on the synthetic route published by Keenan et al.107 The first step was an aldol condensation of commercially available 3-hydroxyacetophenone and 3,4-dimethoxy-benzaldehyde using potassium hydroxide (Fig. 23). The chemoselective reduction of the double bond of 2 was performed in a high pressure autoclave using Lindlar catalyst. The free aromatic alcohol of 3 was subsequently alkylated with tertbutyl 2-bromoacetate. 4a was then subjected to (R)-stereoselective reduction in the autoclave using a Noyori catalyst. 5a was obtained with excellent enantiomeric excess of >95%.

25

Fig 23: (a) KOH, EtOH/H2O, 80-99%, RT. (b) Pd/C/BaSO4, H2 30-40 bar, MeOH, 88-95%. (c) K2CO3, BrCH2COOtBu, aceton, 60-75%. (d) Noyori cat, H2, i-propanol, 80%, >95% ee.

The pipecolinic acid analogs 7a and b were synthesized from commercially available (S)-Pipecolinic respective (S)-thiopipecolinic acid using standard Fmoc protection (Fig. 24). 7a and 7b and the Fmoc-(S)-Proline analog 7c were further esterified with the alcohols 5a and b.

Fig. 24: (a) TEA, Fmoc-Cl, DCM, RT, 90-95%. (b) 5a or 5b EDC, DCM, RT, 50-70%. (c) 4-Methyl-piperidine,

DCM, RT, 70-90%

Based on the FKBP12F36V co-crystal structure with compound 1 it is clear that the Cα substituent has to be in the (S)-configuration to fit best into the new hole (Fig 21)107. First attempts to synthesize different Cα substituted ligands were performed by non-stereoselective alkylation of commercially available 2-(3,4,5 trimethoxyphenyl)acetic acid using various alkylbromides, and eventually separation of the diastereomers of the final product. Unfortunately, the coupling of 17-19 (Fig. 25) to

8 resulted in almost exclusive formation of the unwanted Cα (R)-configuration which was determined by comparison with the HPLC shift of the active isomer of compound 1 and is in line with the low binding affinity of the products 24dia (data not shown). This was subsequently corroborated with the

26

HPLC shifts of the active compound which were later synthesized in a stereoselective manner (compounds 24-28).

We envisioned a stereoselective synthesis using the Evans auxiliary (Fig. 26) to obtain Cα substitutions in (S)-configuration.126 We chose allyl because of its larger size compared to ethyl and because the literature shows broad application of allyl halides in stereoselective Evans alkylation.127, 128 _{Therefore 2-(3,4,5-trimethoxyphenyl)acetic acid 11 was converted to the stable active} pentafluorophenol ester to give 20 which was then coupled with (S)-isopropyloxazolidinone to give the imide 21.

Fig. 25: (a) LiHMDS 2.2 eq, R-Br, THF, RT, 60-80% (b) HATU,

DIEPA, DCM, 50-60%.

The key step in this synthesis, the stereoselective alkylation was performed after formation of the sodium enolate, which reacted with allyl bromide to give 22 with 60% yield and dr >95:5 (determined by HPLC and NMR). The imide was cleaved to give the free acid 23.

Fig. 26: (a) EDC, C6H5OH, DCM, RT 90-95%. (b) BuLi, (S)-isopropyloxazolidinone, THF, -78°C-0°C, 60-80%. (c) NaHMDS, THF, allyl bromide, -78°C, 50-60%. (d) LiOH, H2O2, THF/H2O 8:5, 0°C-RT, 60-90%

27

The allyl substituted ligand series was synthesized by amide bond formation of 8a-c/9a-c with 23 to give the final compounds 24-28 with a dr of >90:10 (determined by HPLC and NMR). Compounds 18 and 19 were obtained after tert-butyl deprotection with 10% TFA in DCM with 50% yield. The 3,4-dehydropipecolinic ester core 33 was synthesized from (S)-allylglycine 29 according to the published procedure of Varray et al.129 33 was coupled to 23 and the ester was cleaved to give 34, which was esterified with the morpholine top group 5b providing the ligand 35 with dr of >95:5.

Fig. 27: (a) HATU, DIPEA, DCM, RT, 40-50%. (b) TFA, DCM, 0°-RT, 50-60%. (c) o-nitrobenzenesulfonyl

chloride, TEA, DCM, RT, 60%. (d) TMSCl, MeOH, 0°C-RT, 99%. (e) Allylbromide, K2CO3, DMF, RT, 83%. (f) RT,Grubbs Cat. II, DCM, reflux, 90%. (g) Thiophenol, Cs2CO2, CH3CN, RT, 81%. (h) HATU, DIPEA, 23, DCM, RT, 68%. (i) LiOH, THF/H2O, RT, 80%. (j) EDC-HCl, DMAP, 5b, DCM, 0°C-RT, 62%.

1.3 Biochemical characterization of the FMSE-ligand-mutant pairs

To determine the IC50 values of the engineered ligands to the proteins, we performed in vitro fluorescence polarization binding assays with the FK1-domains of FKBP51, FKBP52 and the corresponding mutated FK1-domains according to Kozany et al.108 Table 1 shows the binding

28

affinities of the synthesized ligands. We generated a series of Cα allyl compounds bearing different core and top groups. The final compounds were synthesized either with a hydrophilic top group, containing a free acid moiety for better solubility, or a morpholine group for better cell permeability. Compound 24 was the first synthesized and showed already an improvement compared to the ethyl compound 1. The affinity for FKBP51F67V increased upon exchange of ethyl to allyl to 0.7 µM showing that the artificial cavity in the mutated proteins can accommodate bigger substituents than ethyl. We next exchanged the pipecolinic core to proline to probe the effect of a smaller ring on binding. The IC50 of the proline derivative 25 increased by 2-fold for the mutated proteins but also 4-fold for the wildtypes. Thus, the selectivity vs the wild type proteins decreased slightly. By exchanging the hydrophilic acid moiety by morpholine a strong increase in binding affinity occurred. Substitution of the free acid of 24 by morpholine yielded 26 and increased the affinity by 12-fold for the wildtype proteins and by almost 100-fold for the mutated proteins. The selectivity of mutated vs wildtype proteins increased to 1000-fold. Encouraged by this effect we synthesized the proline derivative 27, unfortunately the binding affinity decreased by 3-fold for the wildtype, and 10-fold for the mutant. This is in contrast to proline compound 25 with the free acid top group where this proline modification resulted in an increase of affinity compared to pipecolate. The selectivity of 27 also decreased slightly. We further substituted the core by thiomorpholine 28, which caused a drop of 10-fold in affinity whereas exchange to the 3,4-dehydropipecolinic core 35 showed affinity in the range of 26. Tab. 1 shows the SAR of the allyl series.

Tab. 1: SAR of the allyl series for FKBP51F67V

Fig. 28 shows the binding curves of the two best compounds 26 and 35. Compound 26 was kindly

synthesized by the LDC based on our results as a control for further experiments.

In summary, we improved the binding affinity to the mutated proteins by 1000-fold compared to compound 1 while maintaining 500 to 1000-fold selectivity for mutated FKBP51/52 over wildtype proteins. These selective ligand-mutant pairs can be used in a model system created by chemical genomics to examine the selective inhibition of FKBP51 and 52. This system can be used in different cellular assays (e.g. reporter gene, neurite outgrowth assay) where FKBP mutant proteins can be

29

overexpressed. The best ligands bind with IC50 in the low nanomolar range, which enables for specific inhibition of mutated protein over endogenous proteins.

26

[26](µM)

1e-6 1e-5 1e-4 1e-3 1e-2 1e-1 1e+0 1e+1 1e+2 1e+3

mPs 60 80 100 120 140 160 180 200 220 FKBP51FK1 FKBP52FK1 FKBP51F67V FKBP52F67V

35

1e-6 1e-5 1e-4 1e-3 1e-2 1e-1 1e+0 1e+1 1e+2 1e+3

mPs 60 80 100 120 140 160 180 200 220 FKBP51FK1 FKBP52FK1 FKBP51F67V FKBP52F67V

Fig.28: Biochemical characterization of artificially selective FKBP mutant-ligand

pairs with 26 and 35. Purified FK1-domains of 51wt (2 nM), 52wt (2 nM), 51mut (2 nM) and 52mut (10 nM) were measured in a fluorescence polarization binding assay by titrating 26 or 35 using 3nM of compound F2 (Chapter 4.1) as a tracer108.

30

Tab.2: General structure of the ligands, and binding affinities (IC50) in µM. Purified FK1-domains of 51wt (2 nM), 52wt (2 nM), 51mut (2 nM) and 52mut (10 nM) were measured in a fluorescence polarization binding assay by titrating the compounds using 3nM of compound F2 (Chapter 4.1) as a tracer108. (a) Solubility limit. (b) Compounds were provided by a collaboration with the Lead Discovery Center GmbH (LDC)

31

1.4 Effect of selective inhibition of mutated FKBP51 and 52 on neurite

outgrowth in N2a neuroblastoma cells

FK506 analogs have repeatedly shown neurotrophic or neuroregenerative effects in cellular and animal models. However, due to overlapping functions of FK506-binding proteins and the lack of selectivity of the ligands it proved difficult so far to exactly pinpoint the relevant FKBP. Some FKBPs show negative and others positive effects on neuronal function.

To pharmacologically probe the role of FKBP51 and FKBP52 on neuronal function we used N2a cells as a cellular model for the differentiation of neuronal progenitor cells. Quinta et al. had previously shown that FKBP51 and FKBP52 have opposing effects in this model.86 Similar effects were observed in our system. Overexpressing FKBP51 inhibited neurite outgrowth (Fig. 29B, lanes 3 in Figs. 29C/D) compared to control transfection. In contrast, overexpression of FKBP52 enhanced the outgrowth of neurites compared to control (Fig. 30, lane 3). Fortunately, the neurite outgrowth-suppressing or stimulating effects of FKBP51 or FKBP52 were not affected by the point mutation F67V in the active site.

As we were especially interested in whether FKBP51 can be pharmacologically targeted, we tested the effect of the selective inhibitors on FKBP-modulated neurite outgrowth. Transfection with empty vector pRk5 displays the basal neurite length after starvation (Fig. 29A, Fig. 29D/E, lane 1). The addition of 26 and 35 only marginally affected neurite outgrowth under these conditions. Likewise they did not revert the neurite outgrowth suppression by overexpressed wildtype FKBP51 (Fig 29D/E, lane 4), to which they bind with 1000-fold less affinity than to the mutant FKBP. However, the neurite outgrowth suppressed by the mutated FKBP51 was almost completely rescued by the mutant-selective inhibitors 35 (Fig. 29C, Fig. 29D, lane 6). Almost identical results were obtained with the ligand 26 (Fig. 29E, lane 6).

32 pR K5 ₊ DMS O pR K5 ₊ 35 51_w t + DMS O 51_w t + 35 F6 7V + DMS O F6 7V + 3 5 Neurite length [mm] 0,00 0,02 0,04 0,06 0,08 0,10 pR K5 ₊ DMS O pR K5 ₊₂ 6 51_w t + DMS O 51_w t + 26 51_F6 7V + DMS O 51_F6 7V + 2 6 Ne u ri te le n gth [ mm ] 0,00 0,02 0,04 0,06 0,08 0,10

Fig. 29: Rescue of neurite outgrowth assay in N2a-cells overexpressing FKBP51 by artificially selective ligand 26 and 35. (A)

Transfection control vector prK5 and application of DMSO (B) Overexpression of FKBP51F67V and application of DMSO (C) Overexpression of FKBP51F67V and application of 20 µM Compound 35. (D) and (E) Each bar represents the mean of the neurite length (in mm) of 20-30 cells after the indicated treatment.

Next we tested the consequence of selective inhibition of FKBP52 on neurite outgrowth. Again, 35 did not inhibit the basal neurite outgrowth or the enhancement induced by the wildtype FKBP52 (Fig. 30, lane 2 and 4). In contrast, inhibition of the mutated FKBP52 abolished the neurite stimulation completely even below basal level (Fig. 30, lane 6).

(A)

(B)

(C)

33 pR_k5 + DMS O pR_k5 + 2 6 52_w t + DMS O 52_w t + 26 52_F6 7V + DMS O 52_F6 7V + 2 6 Neu ri te [ mm] 0,00 0,02 0,04 0,06 0,08 0,10

Fig. 30: Suppression of FKBP52 enhanced neurite outgrowth in

N2a-cells overexpressing FKBP52 by artificially selective ligand 35.

The previous results strongly suggested inhibition of FKBP51 and FKBP52 to have opposite effects. To unambiguously proof this we co-overexpressed FKBP51 (wildtype and mutant) and FKBP52 (wildtype and mutant) in all four possible combinations and monitored the effects of the mutant-selective ligand 35 on neurite outgrowth behavior (Fig. 31). Simultaneous overexpression of FKBP51 and 52 wildtype and mutants in all combinations (Fig. 31, lane 3, 5, 7 and 9) resulted in neurite length comparable with basal conditions (overexpression of control vector) which can be attributed to the opposing effects of FKBP51 and FKBP52. As expected, 35 had no effect on basal conditions, and only a small effect on cells overexpressing wildtype FKBP51 and wildtype FKBP52 (Fig. 31, lane 2, 4). Likewise, inhibition of mutated FKBP52 coexpressed with wildtype FKBP51 by 35 caused a shortening of the neurites due to selective, blocking the positive effect of FKBP52 while sparing the suppressing effect of FKBP51 (Fig. 31, lane 7 and 8). Selective inhibition of mutated FKBP51 co-expressed with wildtype FKBP52 by 35 lead to neurite outgrowth in line with leaving the positive effect of FKBP52 (Fig. 31, lane 5 and 6). Importantly inhibiting the two mutated proteins at the same time resulted in neurites with the same length as the corresponding DMSO control (Fig 31 lane 10) consistent with a mutually canceling effect of simultaneously inhibiting both FKBPs. Taken together, in these model experiments with artificially selective ligand mutant pairs, we unambiguously showed that FKBP inhibiting ligands can have neurite outgrowth-stimulating or suppressing effects, depending whether FKBP51 or FKBP52 is more relevant in the system.

In summary, we for the first time demonstrated activity of FKBP51 ligands in a relevant cellular model thereby providing the first experimental proof of concept for the feasibility to

34

pharmacologically target FKBP51. We showed in a cell model of neuronal differentiation, that specific inhibition of overexpressed FKBP51 restores neurite outgrowth whereas specific inhibition of overexpressed FKBP52 has the opposite effect. We therefore propose FKBP51-selective ligands as neuroprotective agents and that selectivity vs. FKBP52 will be crucial for a therapeutic benefit. This could be helpful for neurodegenerative diseases like Alzheimer’s and Parkinson’s disease but also during stress or in depression, which are characterized by neural loss or atrophy.

pRK 5 + DM S_O pRK 5 + 35 51+ 52 + DM S_O 51+ 52 + 35 51* +52 + DM S_O 51* +52 + 35 51+ 52* + _DM S_O 51+ 52* + ₃₅ 51* +52* + _DM S_O 51* +52* + ₃₅ Neurite length [ m m ] 0,00 0,02 0,04 0,06 0,08 0,10

Fig. 31: Co-expression of FKBP51 and 52 and engineered sensitive

mutants thereof. Either addition of DMSO or inhibition by 35. * indicates mutated proteins

35

2. Solving the selectivity issue by an induced fit mechanism

2.1 Induced fit as a basis for selectivity

Fig 32: (A) Co-crystalstructure of SLF and FKBP51FK1. (B) Structure of IF63. (C) Co-crystalstructure of 24 and FKBP51FK1. For

better visibility of the binding pocket K121 was removed from the crystal strucures) highlighted are in blue the important hydrogen bonds from I87 to the C1 carbonyl group and from Y113 to the pipecolinic amide carbonyl C8. Further W90 is shown which displays the bottom of the binding pocket. F67 that is displaced to accommodate 24 is indicated in red.

Our model system (chapter 1.3) showed that selectivity between FKBP51 and FKBP52 is necessary. However, the design of wildtype selective ligands for the wildtype proteins is extremely challenging due to the structural similarity of the binding pocket of the different FKBP subtypes. All ligands tested before showed almost the same binding affinity for FKBP51 and 52.

In the assay results of our model system we noticed a slight preference of some of the Cα-substituted compounds for FKBP51 (1, 24, 25, 26, Tab. 2). This was unexpected as the Cα-allyl group would clash

I87

Y113

W90

F67

F67

W90

I87

_Y113

allyl

(A)

(B)

(C)

(D)

36

with F67 in a binding mode typical for FKBP ligands (Fig. 32A, 34A). Intrigued by this finding we solved the co-crystal structures of 24 at a resolution of 1.4 Å and compared it with the co-crystal structure of the unselective FKBP ligand SLF with FKBP51 (Fig. 32A). This revealed that compound 24 induces a conformational change in the binding pocket of FKBP51 that allows the protruding Cα allyl group to fit into a newly formed hole. More precisely, the Cα allyl substituent of 24 displaces F67 (Fig. 32C) which flips out of the binding pocket to form a new hole in the binding pocket. Fig. 33 shows the superposition of the co-crystal structures of FKBP51FK1/24 and FKBP51FK1/SLF. The major changes are found in the 60s and 70s loop which together with the proline rich loop contains the most important structural differences between the FK1 domains of FKBP51 and FKBP52. The 40s and the proline rich loop are known to be the most flexible part of the protein.109, 110

Fig. 33: Superposition of the backbone traces of the co-crystal structure of FKBP51FK1/24 (pink) and the co-crystal structure of FKBP51FK1/SLF (green).

The most interesting and prominent structural changes are observed in the amino acid sequence from G64 to N74 (GKKFDSSHERN, 60s to 70s loop Fig. 33, Fig. 34). The flip of F67 causes most of the surrounding amino acids to change their conformation. The side chains of K65 and K66 are not defined in the crystal structure, probably due to their flexibility, as they are solvent accessible. No ordered electron density for these amino acids could be observed. Interestingly, D68 almost retains the conformation compared to the co-crystal structure of FKBP51FK1/SLF although it is directly

proline rich

loop

70s loop

F67

51FK1/24

51FK1/SLF

60s loop

37

neighboring F67 and it lost its saltbridge to R73, which possibly is replaced by S70. The conformation of S69 stays the same whereas S70 is flipped upwards together with a large conformational change of H71. The orientation of D72 only slightly changes whereas the orientation of R73 and N74 are completely changed.

Fig. 34: Structural changes of G64 to N74 (60s and 70s) induced by 24. (A) 24 modeled into the co-crystal structure of

FKBP51FK1/SLF. (B) co-crystal structure of FKBP51FK1/24 H71 F67 H71 F67 D68 Y57 D68 Y57 S69 Y57 S70 Y57 D72 Y57 R73 Y57 N74 Y57 K65 F67 G64 K66 G64 K66 K65 F67 S69 Y57 S70 Y57 D72 Y57 N74 Y57 R73 Y57 Cα-Allyl Cα-Allyl

(A)

(B)

clash

38

These extraordinary results provided a structural explanation for the unexpected binding of 24 to FKBP51. We therefore started the synthesis of a series of variously Cα substituted ligands to further elaborate the scope of this induced fit mechanism.

2.2 Synthesis of iFit FKBP51 ligands

The iFit (inducing fit) ligands were synthesized by a similar synthetic route as the FKBP subtype specific engineered ligands.

We synthesized ligands consisting of three main parts the “Core”, the “Cα-Sub” and the “Top group” (Fig. 35). Core structures are either (S)-proline or (S)-pipecolinic acid. The “Top group” is a complex alcohol or amine, containing one or two substituted phenyl/pyridine rings with an ionizable moiety (acid, morpholine or pyridine) to increase solubility or cell permeability. The “Cα-Sub” is an alkyl substituent in Cα-position of the Core with (S)-conformation that protrudes into the hydrophobic binding pocket of FKBP proteins and induces the conformational rearrangement. We synthesized Cα substituents of different sizes ranging from allyl to benzyl to identify the best group for the induced sub-pocket.